1. Field of the Invention
The present invention relates to a grammar update system and method for updating grammar data used for enhancing speech recognition precision in an interactive voice response (IVR) system.
2. Description of the Related Art
In recent years, due to the rapid advancement of a computer environment, such as an enhanced speed of a CPU, an IVR system using an ASR (Auto Speech Recognition) such as a voice portal server, which is conventionally considered to have an excessively heavy operation processing load, is being actively adopted in general applications.
In an IVR system adopted in a conventional application, in order to enhance speech recognition precision, speech recognition processing is generally performed often by using grammar in accordance with the state of interaction. More specifically, grammar is one of the factors that influence the enhancement of precision of speech recognition, the pursuit of ease to use for a user, enlargement of a degree of freedom in utterance, and the like.
For example, in the case of using a so-called dictation technique so as to enhance a degree of freedom in user's utterance, it is required to consider phrases that are hardly uttered. This increases an operation processing amount involved in speech recognition and decreases a speech recognition ratio as a whole, which lowers a level of user's satisfaction and aggravates a time for achieving a task and an achievement percentage.
Thus, there is a strong demand for obtaining finite-state automaton grammar having a scale and complexity sufficient for recognizing user's utterance assumed in accordance with the situation of interaction and applying the grammar to an IVR system.
However, it is practically very difficult to create sufficient grammar data. Therefore, actually, various phrases are assumed and organized in a format of grammar data.
Furthermore, in order to collect a large amount of speech data, the following is often performed: a number of subjects are allowed to use the present system or a pseudo system, whereby a number of phrases are extracted and organized as grammar data.
The following is also performed: speech data and the like in an existing speech interaction system are stored as log data, and a system manager creates new grammar with respect to a portion where misrecognition is caused due to the grammar while extracting the misrecognized portion.
In the case of using the above-mentioned methods, a considerable number of processes are required for creating grammar data. In some cases, it is even difficult to obtain grammar data capable of realizing recognition precision consistent with the number of processes.
For example, regarding dialect, it is required to collect a number of subjects that speak dialect in each area or in a particular area, or to obtain a technical person who sufficiently understands dialect. However, this is practically difficult.
Alternatively, there is a remarkable difference between male and female colloquial expressions. Therefore, a male technical person may not appropriately create grammar data in accordance with a female.